Merge branch 'akpm'
[linux-2.6/next.git] / fs / xfs / xfs_aops.c
blobb90189798d37d104f9d457821c3ea51e4dca2b44
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include "xfs.h"
19 #include "xfs_bit.h"
20 #include "xfs_log.h"
21 #include "xfs_inum.h"
22 #include "xfs_sb.h"
23 #include "xfs_ag.h"
24 #include "xfs_trans.h"
25 #include "xfs_mount.h"
26 #include "xfs_bmap_btree.h"
27 #include "xfs_dinode.h"
28 #include "xfs_inode.h"
29 #include "xfs_alloc.h"
30 #include "xfs_error.h"
31 #include "xfs_rw.h"
32 #include "xfs_iomap.h"
33 #include "xfs_vnodeops.h"
34 #include "xfs_trace.h"
35 #include "xfs_bmap.h"
36 #include <linux/gfp.h>
37 #include <linux/mpage.h>
38 #include <linux/pagevec.h>
39 #include <linux/writeback.h>
43 * Prime number of hash buckets since address is used as the key.
45 #define NVSYNC 37
46 #define to_ioend_wq(v) (&xfs_ioend_wq[((unsigned long)v) % NVSYNC])
47 static wait_queue_head_t xfs_ioend_wq[NVSYNC];
49 void __init
50 xfs_ioend_init(void)
52 int i;
54 for (i = 0; i < NVSYNC; i++)
55 init_waitqueue_head(&xfs_ioend_wq[i]);
58 void
59 xfs_ioend_wait(
60 xfs_inode_t *ip)
62 wait_queue_head_t *wq = to_ioend_wq(ip);
64 wait_event(*wq, (atomic_read(&ip->i_iocount) == 0));
67 STATIC void
68 xfs_ioend_wake(
69 xfs_inode_t *ip)
71 if (atomic_dec_and_test(&ip->i_iocount))
72 wake_up(to_ioend_wq(ip));
75 void
76 xfs_count_page_state(
77 struct page *page,
78 int *delalloc,
79 int *unwritten)
81 struct buffer_head *bh, *head;
83 *delalloc = *unwritten = 0;
85 bh = head = page_buffers(page);
86 do {
87 if (buffer_unwritten(bh))
88 (*unwritten) = 1;
89 else if (buffer_delay(bh))
90 (*delalloc) = 1;
91 } while ((bh = bh->b_this_page) != head);
94 STATIC struct block_device *
95 xfs_find_bdev_for_inode(
96 struct inode *inode)
98 struct xfs_inode *ip = XFS_I(inode);
99 struct xfs_mount *mp = ip->i_mount;
101 if (XFS_IS_REALTIME_INODE(ip))
102 return mp->m_rtdev_targp->bt_bdev;
103 else
104 return mp->m_ddev_targp->bt_bdev;
108 * We're now finished for good with this ioend structure.
109 * Update the page state via the associated buffer_heads,
110 * release holds on the inode and bio, and finally free
111 * up memory. Do not use the ioend after this.
113 STATIC void
114 xfs_destroy_ioend(
115 xfs_ioend_t *ioend)
117 struct buffer_head *bh, *next;
118 struct xfs_inode *ip = XFS_I(ioend->io_inode);
120 for (bh = ioend->io_buffer_head; bh; bh = next) {
121 next = bh->b_private;
122 bh->b_end_io(bh, !ioend->io_error);
126 * Volume managers supporting multiple paths can send back ENODEV
127 * when the final path disappears. In this case continuing to fill
128 * the page cache with dirty data which cannot be written out is
129 * evil, so prevent that.
131 if (unlikely(ioend->io_error == -ENODEV)) {
132 xfs_do_force_shutdown(ip->i_mount, SHUTDOWN_DEVICE_REQ,
133 __FILE__, __LINE__);
136 xfs_ioend_wake(ip);
137 mempool_free(ioend, xfs_ioend_pool);
141 * If the end of the current ioend is beyond the current EOF,
142 * return the new EOF value, otherwise zero.
144 STATIC xfs_fsize_t
145 xfs_ioend_new_eof(
146 xfs_ioend_t *ioend)
148 xfs_inode_t *ip = XFS_I(ioend->io_inode);
149 xfs_fsize_t isize;
150 xfs_fsize_t bsize;
152 bsize = ioend->io_offset + ioend->io_size;
153 isize = MAX(ip->i_size, ip->i_new_size);
154 isize = MIN(isize, bsize);
155 return isize > ip->i_d.di_size ? isize : 0;
159 * Update on-disk file size now that data has been written to disk. The
160 * current in-memory file size is i_size. If a write is beyond eof i_new_size
161 * will be the intended file size until i_size is updated. If this write does
162 * not extend all the way to the valid file size then restrict this update to
163 * the end of the write.
165 * This function does not block as blocking on the inode lock in IO completion
166 * can lead to IO completion order dependency deadlocks.. If it can't get the
167 * inode ilock it will return EAGAIN. Callers must handle this.
169 STATIC int
170 xfs_setfilesize(
171 xfs_ioend_t *ioend)
173 xfs_inode_t *ip = XFS_I(ioend->io_inode);
174 xfs_fsize_t isize;
176 if (unlikely(ioend->io_error))
177 return 0;
179 if (!xfs_ilock_nowait(ip, XFS_ILOCK_EXCL))
180 return EAGAIN;
182 isize = xfs_ioend_new_eof(ioend);
183 if (isize) {
184 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
185 ip->i_d.di_size = isize;
186 xfs_mark_inode_dirty(ip);
189 xfs_iunlock(ip, XFS_ILOCK_EXCL);
190 return 0;
194 * Schedule IO completion handling on the final put of an ioend.
196 STATIC void
197 xfs_finish_ioend(
198 struct xfs_ioend *ioend)
200 if (atomic_dec_and_test(&ioend->io_remaining)) {
201 if (ioend->io_type == IO_UNWRITTEN)
202 queue_work(xfsconvertd_workqueue, &ioend->io_work);
203 else
204 queue_work(xfsdatad_workqueue, &ioend->io_work);
209 * IO write completion.
211 STATIC void
212 xfs_end_io(
213 struct work_struct *work)
215 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
216 struct xfs_inode *ip = XFS_I(ioend->io_inode);
217 int error = 0;
220 * For unwritten extents we need to issue transactions to convert a
221 * range to normal written extens after the data I/O has finished.
223 if (ioend->io_type == IO_UNWRITTEN &&
224 likely(!ioend->io_error && !XFS_FORCED_SHUTDOWN(ip->i_mount))) {
226 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
227 ioend->io_size);
228 if (error)
229 ioend->io_error = error;
233 * We might have to update the on-disk file size after extending
234 * writes.
236 error = xfs_setfilesize(ioend);
237 ASSERT(!error || error == EAGAIN);
240 * If we didn't complete processing of the ioend, requeue it to the
241 * tail of the workqueue for another attempt later. Otherwise destroy
242 * it.
244 if (error == EAGAIN) {
245 atomic_inc(&ioend->io_remaining);
246 xfs_finish_ioend(ioend);
247 /* ensure we don't spin on blocked ioends */
248 delay(1);
249 } else {
250 if (ioend->io_iocb)
251 aio_complete(ioend->io_iocb, ioend->io_result, 0);
252 xfs_destroy_ioend(ioend);
257 * Call IO completion handling in caller context on the final put of an ioend.
259 STATIC void
260 xfs_finish_ioend_sync(
261 struct xfs_ioend *ioend)
263 if (atomic_dec_and_test(&ioend->io_remaining))
264 xfs_end_io(&ioend->io_work);
268 * Allocate and initialise an IO completion structure.
269 * We need to track unwritten extent write completion here initially.
270 * We'll need to extend this for updating the ondisk inode size later
271 * (vs. incore size).
273 STATIC xfs_ioend_t *
274 xfs_alloc_ioend(
275 struct inode *inode,
276 unsigned int type)
278 xfs_ioend_t *ioend;
280 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
283 * Set the count to 1 initially, which will prevent an I/O
284 * completion callback from happening before we have started
285 * all the I/O from calling the completion routine too early.
287 atomic_set(&ioend->io_remaining, 1);
288 ioend->io_error = 0;
289 ioend->io_list = NULL;
290 ioend->io_type = type;
291 ioend->io_inode = inode;
292 ioend->io_buffer_head = NULL;
293 ioend->io_buffer_tail = NULL;
294 atomic_inc(&XFS_I(ioend->io_inode)->i_iocount);
295 ioend->io_offset = 0;
296 ioend->io_size = 0;
297 ioend->io_iocb = NULL;
298 ioend->io_result = 0;
300 INIT_WORK(&ioend->io_work, xfs_end_io);
301 return ioend;
304 STATIC int
305 xfs_map_blocks(
306 struct inode *inode,
307 loff_t offset,
308 struct xfs_bmbt_irec *imap,
309 int type,
310 int nonblocking)
312 struct xfs_inode *ip = XFS_I(inode);
313 struct xfs_mount *mp = ip->i_mount;
314 ssize_t count = 1 << inode->i_blkbits;
315 xfs_fileoff_t offset_fsb, end_fsb;
316 int error = 0;
317 int bmapi_flags = XFS_BMAPI_ENTIRE;
318 int nimaps = 1;
320 if (XFS_FORCED_SHUTDOWN(mp))
321 return -XFS_ERROR(EIO);
323 if (type == IO_UNWRITTEN)
324 bmapi_flags |= XFS_BMAPI_IGSTATE;
326 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
327 if (nonblocking)
328 return -XFS_ERROR(EAGAIN);
329 xfs_ilock(ip, XFS_ILOCK_SHARED);
332 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
333 (ip->i_df.if_flags & XFS_IFEXTENTS));
334 ASSERT(offset <= mp->m_maxioffset);
336 if (offset + count > mp->m_maxioffset)
337 count = mp->m_maxioffset - offset;
338 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
339 offset_fsb = XFS_B_TO_FSBT(mp, offset);
340 error = xfs_bmapi(NULL, ip, offset_fsb, end_fsb - offset_fsb,
341 bmapi_flags, NULL, 0, imap, &nimaps, NULL);
342 xfs_iunlock(ip, XFS_ILOCK_SHARED);
344 if (error)
345 return -XFS_ERROR(error);
347 if (type == IO_DELALLOC &&
348 (!nimaps || isnullstartblock(imap->br_startblock))) {
349 error = xfs_iomap_write_allocate(ip, offset, count, imap);
350 if (!error)
351 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
352 return -XFS_ERROR(error);
355 #ifdef DEBUG
356 if (type == IO_UNWRITTEN) {
357 ASSERT(nimaps);
358 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
359 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
361 #endif
362 if (nimaps)
363 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
364 return 0;
367 STATIC int
368 xfs_imap_valid(
369 struct inode *inode,
370 struct xfs_bmbt_irec *imap,
371 xfs_off_t offset)
373 offset >>= inode->i_blkbits;
375 return offset >= imap->br_startoff &&
376 offset < imap->br_startoff + imap->br_blockcount;
380 * BIO completion handler for buffered IO.
382 STATIC void
383 xfs_end_bio(
384 struct bio *bio,
385 int error)
387 xfs_ioend_t *ioend = bio->bi_private;
389 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
390 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
392 /* Toss bio and pass work off to an xfsdatad thread */
393 bio->bi_private = NULL;
394 bio->bi_end_io = NULL;
395 bio_put(bio);
397 xfs_finish_ioend(ioend);
400 STATIC void
401 xfs_submit_ioend_bio(
402 struct writeback_control *wbc,
403 xfs_ioend_t *ioend,
404 struct bio *bio)
406 atomic_inc(&ioend->io_remaining);
407 bio->bi_private = ioend;
408 bio->bi_end_io = xfs_end_bio;
411 * If the I/O is beyond EOF we mark the inode dirty immediately
412 * but don't update the inode size until I/O completion.
414 if (xfs_ioend_new_eof(ioend))
415 xfs_mark_inode_dirty(XFS_I(ioend->io_inode));
417 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
420 STATIC struct bio *
421 xfs_alloc_ioend_bio(
422 struct buffer_head *bh)
424 int nvecs = bio_get_nr_vecs(bh->b_bdev);
425 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
427 ASSERT(bio->bi_private == NULL);
428 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
429 bio->bi_bdev = bh->b_bdev;
430 return bio;
433 STATIC void
434 xfs_start_buffer_writeback(
435 struct buffer_head *bh)
437 ASSERT(buffer_mapped(bh));
438 ASSERT(buffer_locked(bh));
439 ASSERT(!buffer_delay(bh));
440 ASSERT(!buffer_unwritten(bh));
442 mark_buffer_async_write(bh);
443 set_buffer_uptodate(bh);
444 clear_buffer_dirty(bh);
447 STATIC void
448 xfs_start_page_writeback(
449 struct page *page,
450 int clear_dirty,
451 int buffers)
453 ASSERT(PageLocked(page));
454 ASSERT(!PageWriteback(page));
455 if (clear_dirty)
456 clear_page_dirty_for_io(page);
457 set_page_writeback(page);
458 unlock_page(page);
459 /* If no buffers on the page are to be written, finish it here */
460 if (!buffers)
461 end_page_writeback(page);
464 static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
466 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
470 * Submit all of the bios for all of the ioends we have saved up, covering the
471 * initial writepage page and also any probed pages.
473 * Because we may have multiple ioends spanning a page, we need to start
474 * writeback on all the buffers before we submit them for I/O. If we mark the
475 * buffers as we got, then we can end up with a page that only has buffers
476 * marked async write and I/O complete on can occur before we mark the other
477 * buffers async write.
479 * The end result of this is that we trip a bug in end_page_writeback() because
480 * we call it twice for the one page as the code in end_buffer_async_write()
481 * assumes that all buffers on the page are started at the same time.
483 * The fix is two passes across the ioend list - one to start writeback on the
484 * buffer_heads, and then submit them for I/O on the second pass.
486 STATIC void
487 xfs_submit_ioend(
488 struct writeback_control *wbc,
489 xfs_ioend_t *ioend)
491 xfs_ioend_t *head = ioend;
492 xfs_ioend_t *next;
493 struct buffer_head *bh;
494 struct bio *bio;
495 sector_t lastblock = 0;
497 /* Pass 1 - start writeback */
498 do {
499 next = ioend->io_list;
500 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
501 xfs_start_buffer_writeback(bh);
502 } while ((ioend = next) != NULL);
504 /* Pass 2 - submit I/O */
505 ioend = head;
506 do {
507 next = ioend->io_list;
508 bio = NULL;
510 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
512 if (!bio) {
513 retry:
514 bio = xfs_alloc_ioend_bio(bh);
515 } else if (bh->b_blocknr != lastblock + 1) {
516 xfs_submit_ioend_bio(wbc, ioend, bio);
517 goto retry;
520 if (bio_add_buffer(bio, bh) != bh->b_size) {
521 xfs_submit_ioend_bio(wbc, ioend, bio);
522 goto retry;
525 lastblock = bh->b_blocknr;
527 if (bio)
528 xfs_submit_ioend_bio(wbc, ioend, bio);
529 xfs_finish_ioend(ioend);
530 } while ((ioend = next) != NULL);
534 * Cancel submission of all buffer_heads so far in this endio.
535 * Toss the endio too. Only ever called for the initial page
536 * in a writepage request, so only ever one page.
538 STATIC void
539 xfs_cancel_ioend(
540 xfs_ioend_t *ioend)
542 xfs_ioend_t *next;
543 struct buffer_head *bh, *next_bh;
545 do {
546 next = ioend->io_list;
547 bh = ioend->io_buffer_head;
548 do {
549 next_bh = bh->b_private;
550 clear_buffer_async_write(bh);
551 unlock_buffer(bh);
552 } while ((bh = next_bh) != NULL);
554 xfs_ioend_wake(XFS_I(ioend->io_inode));
555 mempool_free(ioend, xfs_ioend_pool);
556 } while ((ioend = next) != NULL);
560 * Test to see if we've been building up a completion structure for
561 * earlier buffers -- if so, we try to append to this ioend if we
562 * can, otherwise we finish off any current ioend and start another.
563 * Return true if we've finished the given ioend.
565 STATIC void
566 xfs_add_to_ioend(
567 struct inode *inode,
568 struct buffer_head *bh,
569 xfs_off_t offset,
570 unsigned int type,
571 xfs_ioend_t **result,
572 int need_ioend)
574 xfs_ioend_t *ioend = *result;
576 if (!ioend || need_ioend || type != ioend->io_type) {
577 xfs_ioend_t *previous = *result;
579 ioend = xfs_alloc_ioend(inode, type);
580 ioend->io_offset = offset;
581 ioend->io_buffer_head = bh;
582 ioend->io_buffer_tail = bh;
583 if (previous)
584 previous->io_list = ioend;
585 *result = ioend;
586 } else {
587 ioend->io_buffer_tail->b_private = bh;
588 ioend->io_buffer_tail = bh;
591 bh->b_private = NULL;
592 ioend->io_size += bh->b_size;
595 STATIC void
596 xfs_map_buffer(
597 struct inode *inode,
598 struct buffer_head *bh,
599 struct xfs_bmbt_irec *imap,
600 xfs_off_t offset)
602 sector_t bn;
603 struct xfs_mount *m = XFS_I(inode)->i_mount;
604 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
605 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
607 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
608 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
610 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
611 ((offset - iomap_offset) >> inode->i_blkbits);
613 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
615 bh->b_blocknr = bn;
616 set_buffer_mapped(bh);
619 STATIC void
620 xfs_map_at_offset(
621 struct inode *inode,
622 struct buffer_head *bh,
623 struct xfs_bmbt_irec *imap,
624 xfs_off_t offset)
626 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
627 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
629 xfs_map_buffer(inode, bh, imap, offset);
630 set_buffer_mapped(bh);
631 clear_buffer_delay(bh);
632 clear_buffer_unwritten(bh);
636 * Test if a given page is suitable for writing as part of an unwritten
637 * or delayed allocate extent.
639 STATIC int
640 xfs_is_delayed_page(
641 struct page *page,
642 unsigned int type)
644 if (PageWriteback(page))
645 return 0;
647 if (page->mapping && page_has_buffers(page)) {
648 struct buffer_head *bh, *head;
649 int acceptable = 0;
651 bh = head = page_buffers(page);
652 do {
653 if (buffer_unwritten(bh))
654 acceptable = (type == IO_UNWRITTEN);
655 else if (buffer_delay(bh))
656 acceptable = (type == IO_DELALLOC);
657 else if (buffer_dirty(bh) && buffer_mapped(bh))
658 acceptable = (type == IO_OVERWRITE);
659 else
660 break;
661 } while ((bh = bh->b_this_page) != head);
663 if (acceptable)
664 return 1;
667 return 0;
671 * Allocate & map buffers for page given the extent map. Write it out.
672 * except for the original page of a writepage, this is called on
673 * delalloc/unwritten pages only, for the original page it is possible
674 * that the page has no mapping at all.
676 STATIC int
677 xfs_convert_page(
678 struct inode *inode,
679 struct page *page,
680 loff_t tindex,
681 struct xfs_bmbt_irec *imap,
682 xfs_ioend_t **ioendp,
683 struct writeback_control *wbc)
685 struct buffer_head *bh, *head;
686 xfs_off_t end_offset;
687 unsigned long p_offset;
688 unsigned int type;
689 int len, page_dirty;
690 int count = 0, done = 0, uptodate = 1;
691 xfs_off_t offset = page_offset(page);
693 if (page->index != tindex)
694 goto fail;
695 if (!trylock_page(page))
696 goto fail;
697 if (PageWriteback(page))
698 goto fail_unlock_page;
699 if (page->mapping != inode->i_mapping)
700 goto fail_unlock_page;
701 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
702 goto fail_unlock_page;
705 * page_dirty is initially a count of buffers on the page before
706 * EOF and is decremented as we move each into a cleanable state.
708 * Derivation:
710 * End offset is the highest offset that this page should represent.
711 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
712 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
713 * hence give us the correct page_dirty count. On any other page,
714 * it will be zero and in that case we need page_dirty to be the
715 * count of buffers on the page.
717 end_offset = min_t(unsigned long long,
718 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
719 i_size_read(inode));
721 len = 1 << inode->i_blkbits;
722 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
723 PAGE_CACHE_SIZE);
724 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
725 page_dirty = p_offset / len;
727 bh = head = page_buffers(page);
728 do {
729 if (offset >= end_offset)
730 break;
731 if (!buffer_uptodate(bh))
732 uptodate = 0;
733 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
734 done = 1;
735 continue;
738 if (buffer_unwritten(bh) || buffer_delay(bh) ||
739 buffer_mapped(bh)) {
740 if (buffer_unwritten(bh))
741 type = IO_UNWRITTEN;
742 else if (buffer_delay(bh))
743 type = IO_DELALLOC;
744 else
745 type = IO_OVERWRITE;
747 if (!xfs_imap_valid(inode, imap, offset)) {
748 done = 1;
749 continue;
752 lock_buffer(bh);
753 if (type != IO_OVERWRITE)
754 xfs_map_at_offset(inode, bh, imap, offset);
755 xfs_add_to_ioend(inode, bh, offset, type,
756 ioendp, done);
758 page_dirty--;
759 count++;
760 } else {
761 done = 1;
763 } while (offset += len, (bh = bh->b_this_page) != head);
765 if (uptodate && bh == head)
766 SetPageUptodate(page);
768 if (count) {
769 if (--wbc->nr_to_write <= 0 &&
770 wbc->sync_mode == WB_SYNC_NONE)
771 done = 1;
773 xfs_start_page_writeback(page, !page_dirty, count);
775 return done;
776 fail_unlock_page:
777 unlock_page(page);
778 fail:
779 return 1;
783 * Convert & write out a cluster of pages in the same extent as defined
784 * by mp and following the start page.
786 STATIC void
787 xfs_cluster_write(
788 struct inode *inode,
789 pgoff_t tindex,
790 struct xfs_bmbt_irec *imap,
791 xfs_ioend_t **ioendp,
792 struct writeback_control *wbc,
793 pgoff_t tlast)
795 struct pagevec pvec;
796 int done = 0, i;
798 pagevec_init(&pvec, 0);
799 while (!done && tindex <= tlast) {
800 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
802 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
803 break;
805 for (i = 0; i < pagevec_count(&pvec); i++) {
806 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
807 imap, ioendp, wbc);
808 if (done)
809 break;
812 pagevec_release(&pvec);
813 cond_resched();
817 STATIC void
818 xfs_vm_invalidatepage(
819 struct page *page,
820 unsigned long offset)
822 trace_xfs_invalidatepage(page->mapping->host, page, offset);
823 block_invalidatepage(page, offset);
827 * If the page has delalloc buffers on it, we need to punch them out before we
828 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
829 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
830 * is done on that same region - the delalloc extent is returned when none is
831 * supposed to be there.
833 * We prevent this by truncating away the delalloc regions on the page before
834 * invalidating it. Because they are delalloc, we can do this without needing a
835 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
836 * truncation without a transaction as there is no space left for block
837 * reservation (typically why we see a ENOSPC in writeback).
839 * This is not a performance critical path, so for now just do the punching a
840 * buffer head at a time.
842 STATIC void
843 xfs_aops_discard_page(
844 struct page *page)
846 struct inode *inode = page->mapping->host;
847 struct xfs_inode *ip = XFS_I(inode);
848 struct buffer_head *bh, *head;
849 loff_t offset = page_offset(page);
851 if (!xfs_is_delayed_page(page, IO_DELALLOC))
852 goto out_invalidate;
854 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
855 goto out_invalidate;
857 xfs_alert(ip->i_mount,
858 "page discard on page %p, inode 0x%llx, offset %llu.",
859 page, ip->i_ino, offset);
861 xfs_ilock(ip, XFS_ILOCK_EXCL);
862 bh = head = page_buffers(page);
863 do {
864 int error;
865 xfs_fileoff_t start_fsb;
867 if (!buffer_delay(bh))
868 goto next_buffer;
870 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
871 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
872 if (error) {
873 /* something screwed, just bail */
874 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
875 xfs_alert(ip->i_mount,
876 "page discard unable to remove delalloc mapping.");
878 break;
880 next_buffer:
881 offset += 1 << inode->i_blkbits;
883 } while ((bh = bh->b_this_page) != head);
885 xfs_iunlock(ip, XFS_ILOCK_EXCL);
886 out_invalidate:
887 xfs_vm_invalidatepage(page, 0);
888 return;
892 * Write out a dirty page.
894 * For delalloc space on the page we need to allocate space and flush it.
895 * For unwritten space on the page we need to start the conversion to
896 * regular allocated space.
897 * For any other dirty buffer heads on the page we should flush them.
899 STATIC int
900 xfs_vm_writepage(
901 struct page *page,
902 struct writeback_control *wbc)
904 struct inode *inode = page->mapping->host;
905 struct buffer_head *bh, *head;
906 struct xfs_bmbt_irec imap;
907 xfs_ioend_t *ioend = NULL, *iohead = NULL;
908 loff_t offset;
909 unsigned int type;
910 __uint64_t end_offset;
911 pgoff_t end_index, last_index;
912 ssize_t len;
913 int err, imap_valid = 0, uptodate = 1;
914 int count = 0;
915 int nonblocking = 0;
917 trace_xfs_writepage(inode, page, 0);
919 ASSERT(page_has_buffers(page));
922 * Refuse to write the page out if we are called from reclaim context.
924 * This avoids stack overflows when called from deeply used stacks in
925 * random callers for direct reclaim or memcg reclaim. We explicitly
926 * allow reclaim from kswapd as the stack usage there is relatively low.
928 * This should never happen except in the case of a VM regression so
929 * warn about it.
931 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
932 PF_MEMALLOC))
933 goto redirty;
936 * Given that we do not allow direct reclaim to call us, we should
937 * never be called while in a filesystem transaction.
939 if (WARN_ON(current->flags & PF_FSTRANS))
940 goto redirty;
942 /* Is this page beyond the end of the file? */
943 offset = i_size_read(inode);
944 end_index = offset >> PAGE_CACHE_SHIFT;
945 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
946 if (page->index >= end_index) {
947 if ((page->index >= end_index + 1) ||
948 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
949 unlock_page(page);
950 return 0;
954 end_offset = min_t(unsigned long long,
955 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
956 offset);
957 len = 1 << inode->i_blkbits;
959 bh = head = page_buffers(page);
960 offset = page_offset(page);
961 type = IO_OVERWRITE;
963 if (wbc->sync_mode == WB_SYNC_NONE)
964 nonblocking = 1;
966 do {
967 int new_ioend = 0;
969 if (offset >= end_offset)
970 break;
971 if (!buffer_uptodate(bh))
972 uptodate = 0;
975 * set_page_dirty dirties all buffers in a page, independent
976 * of their state. The dirty state however is entirely
977 * meaningless for holes (!mapped && uptodate), so skip
978 * buffers covering holes here.
980 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
981 imap_valid = 0;
982 continue;
985 if (buffer_unwritten(bh)) {
986 if (type != IO_UNWRITTEN) {
987 type = IO_UNWRITTEN;
988 imap_valid = 0;
990 } else if (buffer_delay(bh)) {
991 if (type != IO_DELALLOC) {
992 type = IO_DELALLOC;
993 imap_valid = 0;
995 } else if (buffer_uptodate(bh)) {
996 if (type != IO_OVERWRITE) {
997 type = IO_OVERWRITE;
998 imap_valid = 0;
1000 } else {
1001 if (PageUptodate(page)) {
1002 ASSERT(buffer_mapped(bh));
1003 imap_valid = 0;
1005 continue;
1008 if (imap_valid)
1009 imap_valid = xfs_imap_valid(inode, &imap, offset);
1010 if (!imap_valid) {
1012 * If we didn't have a valid mapping then we need to
1013 * put the new mapping into a separate ioend structure.
1014 * This ensures non-contiguous extents always have
1015 * separate ioends, which is particularly important
1016 * for unwritten extent conversion at I/O completion
1017 * time.
1019 new_ioend = 1;
1020 err = xfs_map_blocks(inode, offset, &imap, type,
1021 nonblocking);
1022 if (err)
1023 goto error;
1024 imap_valid = xfs_imap_valid(inode, &imap, offset);
1026 if (imap_valid) {
1027 lock_buffer(bh);
1028 if (type != IO_OVERWRITE)
1029 xfs_map_at_offset(inode, bh, &imap, offset);
1030 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1031 new_ioend);
1032 count++;
1035 if (!iohead)
1036 iohead = ioend;
1038 } while (offset += len, ((bh = bh->b_this_page) != head));
1040 if (uptodate && bh == head)
1041 SetPageUptodate(page);
1043 xfs_start_page_writeback(page, 1, count);
1045 if (ioend && imap_valid) {
1046 xfs_off_t end_index;
1048 end_index = imap.br_startoff + imap.br_blockcount;
1050 /* to bytes */
1051 end_index <<= inode->i_blkbits;
1053 /* to pages */
1054 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1056 /* check against file size */
1057 if (end_index > last_index)
1058 end_index = last_index;
1060 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
1061 wbc, end_index);
1064 if (iohead)
1065 xfs_submit_ioend(wbc, iohead);
1067 return 0;
1069 error:
1070 if (iohead)
1071 xfs_cancel_ioend(iohead);
1073 if (err == -EAGAIN)
1074 goto redirty;
1076 xfs_aops_discard_page(page);
1077 ClearPageUptodate(page);
1078 unlock_page(page);
1079 return err;
1081 redirty:
1082 redirty_page_for_writepage(wbc, page);
1083 unlock_page(page);
1084 return 0;
1087 STATIC int
1088 xfs_vm_writepages(
1089 struct address_space *mapping,
1090 struct writeback_control *wbc)
1092 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1093 return generic_writepages(mapping, wbc);
1097 * Called to move a page into cleanable state - and from there
1098 * to be released. The page should already be clean. We always
1099 * have buffer heads in this call.
1101 * Returns 1 if the page is ok to release, 0 otherwise.
1103 STATIC int
1104 xfs_vm_releasepage(
1105 struct page *page,
1106 gfp_t gfp_mask)
1108 int delalloc, unwritten;
1110 trace_xfs_releasepage(page->mapping->host, page, 0);
1112 xfs_count_page_state(page, &delalloc, &unwritten);
1114 if (WARN_ON(delalloc))
1115 return 0;
1116 if (WARN_ON(unwritten))
1117 return 0;
1119 return try_to_free_buffers(page);
1122 STATIC int
1123 __xfs_get_blocks(
1124 struct inode *inode,
1125 sector_t iblock,
1126 struct buffer_head *bh_result,
1127 int create,
1128 int direct)
1130 struct xfs_inode *ip = XFS_I(inode);
1131 struct xfs_mount *mp = ip->i_mount;
1132 xfs_fileoff_t offset_fsb, end_fsb;
1133 int error = 0;
1134 int lockmode = 0;
1135 struct xfs_bmbt_irec imap;
1136 int nimaps = 1;
1137 xfs_off_t offset;
1138 ssize_t size;
1139 int new = 0;
1141 if (XFS_FORCED_SHUTDOWN(mp))
1142 return -XFS_ERROR(EIO);
1144 offset = (xfs_off_t)iblock << inode->i_blkbits;
1145 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1146 size = bh_result->b_size;
1148 if (!create && direct && offset >= i_size_read(inode))
1149 return 0;
1151 if (create) {
1152 lockmode = XFS_ILOCK_EXCL;
1153 xfs_ilock(ip, lockmode);
1154 } else {
1155 lockmode = xfs_ilock_map_shared(ip);
1158 ASSERT(offset <= mp->m_maxioffset);
1159 if (offset + size > mp->m_maxioffset)
1160 size = mp->m_maxioffset - offset;
1161 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1162 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1164 error = xfs_bmapi(NULL, ip, offset_fsb, end_fsb - offset_fsb,
1165 XFS_BMAPI_ENTIRE, NULL, 0, &imap, &nimaps, NULL);
1166 if (error)
1167 goto out_unlock;
1169 if (create &&
1170 (!nimaps ||
1171 (imap.br_startblock == HOLESTARTBLOCK ||
1172 imap.br_startblock == DELAYSTARTBLOCK))) {
1173 if (direct) {
1174 error = xfs_iomap_write_direct(ip, offset, size,
1175 &imap, nimaps);
1176 } else {
1177 error = xfs_iomap_write_delay(ip, offset, size, &imap);
1179 if (error)
1180 goto out_unlock;
1182 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1183 } else if (nimaps) {
1184 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
1185 } else {
1186 trace_xfs_get_blocks_notfound(ip, offset, size);
1187 goto out_unlock;
1189 xfs_iunlock(ip, lockmode);
1191 if (imap.br_startblock != HOLESTARTBLOCK &&
1192 imap.br_startblock != DELAYSTARTBLOCK) {
1194 * For unwritten extents do not report a disk address on
1195 * the read case (treat as if we're reading into a hole).
1197 if (create || !ISUNWRITTEN(&imap))
1198 xfs_map_buffer(inode, bh_result, &imap, offset);
1199 if (create && ISUNWRITTEN(&imap)) {
1200 if (direct)
1201 bh_result->b_private = inode;
1202 set_buffer_unwritten(bh_result);
1207 * If this is a realtime file, data may be on a different device.
1208 * to that pointed to from the buffer_head b_bdev currently.
1210 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1213 * If we previously allocated a block out beyond eof and we are now
1214 * coming back to use it then we will need to flag it as new even if it
1215 * has a disk address.
1217 * With sub-block writes into unwritten extents we also need to mark
1218 * the buffer as new so that the unwritten parts of the buffer gets
1219 * correctly zeroed.
1221 if (create &&
1222 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
1223 (offset >= i_size_read(inode)) ||
1224 (new || ISUNWRITTEN(&imap))))
1225 set_buffer_new(bh_result);
1227 if (imap.br_startblock == DELAYSTARTBLOCK) {
1228 BUG_ON(direct);
1229 if (create) {
1230 set_buffer_uptodate(bh_result);
1231 set_buffer_mapped(bh_result);
1232 set_buffer_delay(bh_result);
1237 * If this is O_DIRECT or the mpage code calling tell them how large
1238 * the mapping is, so that we can avoid repeated get_blocks calls.
1240 if (direct || size > (1 << inode->i_blkbits)) {
1241 xfs_off_t mapping_size;
1243 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1244 mapping_size <<= inode->i_blkbits;
1246 ASSERT(mapping_size > 0);
1247 if (mapping_size > size)
1248 mapping_size = size;
1249 if (mapping_size > LONG_MAX)
1250 mapping_size = LONG_MAX;
1252 bh_result->b_size = mapping_size;
1255 return 0;
1257 out_unlock:
1258 xfs_iunlock(ip, lockmode);
1259 return -error;
1263 xfs_get_blocks(
1264 struct inode *inode,
1265 sector_t iblock,
1266 struct buffer_head *bh_result,
1267 int create)
1269 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
1272 STATIC int
1273 xfs_get_blocks_direct(
1274 struct inode *inode,
1275 sector_t iblock,
1276 struct buffer_head *bh_result,
1277 int create)
1279 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
1283 * Complete a direct I/O write request.
1285 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1286 * need to issue a transaction to convert the range from unwritten to written
1287 * extents. In case this is regular synchronous I/O we just call xfs_end_io
1288 * to do this and we are done. But in case this was a successful AIO
1289 * request this handler is called from interrupt context, from which we
1290 * can't start transactions. In that case offload the I/O completion to
1291 * the workqueues we also use for buffered I/O completion.
1293 STATIC void
1294 xfs_end_io_direct_write(
1295 struct kiocb *iocb,
1296 loff_t offset,
1297 ssize_t size,
1298 void *private,
1299 int ret,
1300 bool is_async)
1302 struct xfs_ioend *ioend = iocb->private;
1305 * blockdev_direct_IO can return an error even after the I/O
1306 * completion handler was called. Thus we need to protect
1307 * against double-freeing.
1309 iocb->private = NULL;
1311 ioend->io_offset = offset;
1312 ioend->io_size = size;
1313 if (private && size > 0)
1314 ioend->io_type = IO_UNWRITTEN;
1316 if (is_async) {
1318 * If we are converting an unwritten extent we need to delay
1319 * the AIO completion until after the unwrittent extent
1320 * conversion has completed, otherwise do it ASAP.
1322 if (ioend->io_type == IO_UNWRITTEN) {
1323 ioend->io_iocb = iocb;
1324 ioend->io_result = ret;
1325 } else {
1326 aio_complete(iocb, ret, 0);
1328 xfs_finish_ioend(ioend);
1329 } else {
1330 xfs_finish_ioend_sync(ioend);
1333 /* XXX: probably should move into the real I/O completion handler */
1334 inode_dio_done(ioend->io_inode);
1337 STATIC ssize_t
1338 xfs_vm_direct_IO(
1339 int rw,
1340 struct kiocb *iocb,
1341 const struct iovec *iov,
1342 loff_t offset,
1343 unsigned long nr_segs)
1345 struct inode *inode = iocb->ki_filp->f_mapping->host;
1346 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
1347 ssize_t ret;
1349 if (rw & WRITE) {
1350 iocb->private = xfs_alloc_ioend(inode, IO_DIRECT);
1352 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1353 offset, nr_segs,
1354 xfs_get_blocks_direct,
1355 xfs_end_io_direct_write, NULL, 0);
1356 if (ret != -EIOCBQUEUED && iocb->private)
1357 xfs_destroy_ioend(iocb->private);
1358 } else {
1359 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1360 offset, nr_segs,
1361 xfs_get_blocks_direct,
1362 NULL, NULL, 0);
1365 return ret;
1368 STATIC void
1369 xfs_vm_write_failed(
1370 struct address_space *mapping,
1371 loff_t to)
1373 struct inode *inode = mapping->host;
1375 if (to > inode->i_size) {
1377 * punch out the delalloc blocks we have already allocated. We
1378 * don't call xfs_setattr() to do this as we may be in the
1379 * middle of a multi-iovec write and so the vfs inode->i_size
1380 * will not match the xfs ip->i_size and so it will zero too
1381 * much. Hence we jus truncate the page cache to zero what is
1382 * necessary and punch the delalloc blocks directly.
1384 struct xfs_inode *ip = XFS_I(inode);
1385 xfs_fileoff_t start_fsb;
1386 xfs_fileoff_t end_fsb;
1387 int error;
1389 truncate_pagecache(inode, to, inode->i_size);
1392 * Check if there are any blocks that are outside of i_size
1393 * that need to be trimmed back.
1395 start_fsb = XFS_B_TO_FSB(ip->i_mount, inode->i_size) + 1;
1396 end_fsb = XFS_B_TO_FSB(ip->i_mount, to);
1397 if (end_fsb <= start_fsb)
1398 return;
1400 xfs_ilock(ip, XFS_ILOCK_EXCL);
1401 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1402 end_fsb - start_fsb);
1403 if (error) {
1404 /* something screwed, just bail */
1405 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1406 xfs_alert(ip->i_mount,
1407 "xfs_vm_write_failed: unable to clean up ino %lld",
1408 ip->i_ino);
1411 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1415 STATIC int
1416 xfs_vm_write_begin(
1417 struct file *file,
1418 struct address_space *mapping,
1419 loff_t pos,
1420 unsigned len,
1421 unsigned flags,
1422 struct page **pagep,
1423 void **fsdata)
1425 int ret;
1427 ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS,
1428 pagep, xfs_get_blocks);
1429 if (unlikely(ret))
1430 xfs_vm_write_failed(mapping, pos + len);
1431 return ret;
1434 STATIC int
1435 xfs_vm_write_end(
1436 struct file *file,
1437 struct address_space *mapping,
1438 loff_t pos,
1439 unsigned len,
1440 unsigned copied,
1441 struct page *page,
1442 void *fsdata)
1444 int ret;
1446 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
1447 if (unlikely(ret < len))
1448 xfs_vm_write_failed(mapping, pos + len);
1449 return ret;
1452 STATIC sector_t
1453 xfs_vm_bmap(
1454 struct address_space *mapping,
1455 sector_t block)
1457 struct inode *inode = (struct inode *)mapping->host;
1458 struct xfs_inode *ip = XFS_I(inode);
1460 trace_xfs_vm_bmap(XFS_I(inode));
1461 xfs_ilock(ip, XFS_IOLOCK_SHARED);
1462 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
1463 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
1464 return generic_block_bmap(mapping, block, xfs_get_blocks);
1467 STATIC int
1468 xfs_vm_readpage(
1469 struct file *unused,
1470 struct page *page)
1472 return mpage_readpage(page, xfs_get_blocks);
1475 STATIC int
1476 xfs_vm_readpages(
1477 struct file *unused,
1478 struct address_space *mapping,
1479 struct list_head *pages,
1480 unsigned nr_pages)
1482 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1485 const struct address_space_operations xfs_address_space_operations = {
1486 .readpage = xfs_vm_readpage,
1487 .readpages = xfs_vm_readpages,
1488 .writepage = xfs_vm_writepage,
1489 .writepages = xfs_vm_writepages,
1490 .releasepage = xfs_vm_releasepage,
1491 .invalidatepage = xfs_vm_invalidatepage,
1492 .write_begin = xfs_vm_write_begin,
1493 .write_end = xfs_vm_write_end,
1494 .bmap = xfs_vm_bmap,
1495 .direct_IO = xfs_vm_direct_IO,
1496 .migratepage = buffer_migrate_page,
1497 .is_partially_uptodate = block_is_partially_uptodate,
1498 .error_remove_page = generic_error_remove_page,